[JJenkx]

What

A host directory-creation failure currently throws across the guest ABI. A block-count write can also fail after the mount alias, slot and result have already been published. Return the existing SAVE_DATA_ERROR_INTERNAL for filesystem failures, persist the block-count sidecar through the existing replacement helper before publishing the mount, and remove a newly created empty save directory when size persistence fails. Filesystem probes use std::error_code overloads, including mounted-directory identity checks, so ordinary host failures do not depend on exception unwinding through the Windows PRX. Missing directories retain the normal create/open behavior. Failed calls preserve the caller's result and any existing alias; traces inspect mount output only on success.

The regression checks a regular file blocking the save root and a directory blocking size-metadata replacement, including slot/alias/result preservation and recovery. The per-host-error correspondence to PS5 errors remains unverified and is recorded in library debt.

Tested

Linux 7.2.9-arch1-1, c++ (GCC) 16.2.1 20260810. Pinned main 1c6b316cf762c7a8d34623907d62bc1cba2217b6 versus 2606c33532a52416ed993f56814441e7d80f1a96.

Rebuilt the complete SaveData.native module and test sources using the existing Release compile/link recipes in independent outputs. Both arms use identical SHA-pinned frozen kernel/libc dependencies; this is module/suite qualification, not a clean whole-repository build.

• guest_savedata_search, guest_savedata_delete, guest_savedata_convert, guest_savedata_mount_status, guest_savedata_ps4, guest_savedata_cancel: three runs each on each arm, 18/18 baseline and 18/18 change passes, no whole-test skips.
• Main module with the change's tests: expected failure in the new host-error/state-preservation checks.
• Real unprivileged jjenkx process against a private root-owned save directory: baseline aborts 3/3 with Permission denied; change passes 3/3, returning INTERNAL while preserving the result and leaving no mounted slot.
• Conventions and whitespace checks pass. No macOS execution, game success, GPU result or firmware error-mapping measurement is claimed. Windows execution is recorded below.

The unmodified overlap checker scanned all 1,444 open PRs against main 58d5af7daa8087d2ffbdb50ae6e0db3cb6a88e48; it reported no completed positive overlap. It could not compare #10674 because that branch conflicts with main. Its shared diff only changes the sceSaveDataGetProgress(float*) signature and corresponding debt, independent of Mount3. A private source merge with #12748 also passes after avoiding unrelated test include/constant conflicts; no combined runtime qualification is claimed. This branch merges cleanly with checked main 58d5af7daa8087d2ffbdb50ae6e0db3cb6a88e48.

Reproduction recipes, synthetic probe, full suite logs, negative controls and overlap evidence, SHA-256 eb8db3423918202bceaa39eb8ba550d1bf469d715d9f981c6ddeedfc7dd12621 (download verified).
Follow-up 2606c33532a52416ed993f56814441e7d80f1a96 addresses the Windows CI failure reported in review: the previous catch could not receive a filesystem exception through the PRX. The catch is removed and host probes return errors directly. The existing block-size replacement helper already uses error-code directory creation and native/nonthrowing rename operations; blocks_path only constructs the sidecar name. The Linux suite and unprivileged permission results above were rerun against this exact follow-up, with identical counts; conventions pass against fetched main 4bc9e779aa7abb2109c65ee3bca19e1b85df3be8, and a private merge-tree check against that main is clean. The Windows rerun completed successfully; see the exact-head comparison below.

Follow-up source recipes, synthetic test logs, negative control and clean merge receipt, SHA-256 98a946c01c518e72f0081fabed3036e7f9664354c49b78c147cccc90afb85cf8 (download verified). The earlier archive above covers the prior head only.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: none
• Unblocks: none (host-I/O error handling)
• GPUs tested: n/a
• AI-assisted: yes (Codex)

Windows validation of the follow-up

On Windows 11 x86-64 with WinLibs GCC 15.2 UCRT r7, built the SaveData module/dependencies and ran the six actual guest SaveData suites in one isolated CMake build:

• Previous head 8086a37957a2ad182ae9562e10e24687edc2947e: 5/6 pass; guest_savedata_mount_status aborts with filesystem_error and libSceSaveData.native.prx ... has no unwind data, reproducing the review failure.
• Follow-up 2606c33532a52416ed993f56814441e7d80f1a96: 6/6 pass, zero skips. Search, delete, convert, mount status, PS4 and cancel all execute.

Windows build commands, full CTest logs, JUnit results and module hashes, SHA-256 75e001074bd44a5c82564673b1fba0b791532074a6b32ff5beda9149e06c297e (download verified). This is focused module/suite qualification, not a whole-project build or game/GPU test. The first setup attempt stopped before compilation because the donor lacked a dependency; the fresh second attempt checked out all 31 pinned dependencies and produced these results.

[MarcZX10]

Reviewed — looks good

The mount path now converts filesystem failures to the existing internal error before publishing an alias, slot, or result, and the supplied Linux evidence exercises those failure paths.

What I ran

Reviewed head 8086a37957a2ad182ae9562e10e24687edc2947e against base f91137142eafd5df646fca6419a1fc4ef61b89d9; the merge preview was clean (tree e0a3bd33285606d39994c8a5f3b154df986e87dd). Environment: Debian 12 x86-64, Python 3.11.2, Git 2.39.5. I did not compile or run C++ tests locally.

python3 tools/pr_precheck.py --repo boykopovar/AnyPS5 --pr 13324
python3 tools/pr_abi_audit.py --repo /marc/AnyPS5 --base f91137142eafd5df646fca6419a1fc4ef61b89d9 --head 8086a37957a2ad182ae9562e10e24687edc2947e --json
python3 tools/pr_test_quality.py --root /marc/AnyPS5 --json core/libs/prx/libSceSaveData.native/Export.cpp core/libs/tests/GuestSaveDataMountStatus.cpp docs/dev/debt/libSceSaveData.native.md
git diff --check f91137142eafd5df646fca6419a1fc4ef61b89d9...8086a37957a2ad182ae9562e10e24687edc2947e
python3 tools/check_conventions.py --base f91137142eafd5df646fca6419a1fc4ef61b89d9 --head 8086a37957a2ad182ae9562e10e24687edc2947e --title 'fix(savedata): keep failed host mounts unpublished' --body /tmp/anyps5-pr-precheck/boykopovar-AnyPS5-pr-13324-8086a37957a2/pr-body.md
git merge-tree --write-tree f91137142eafd5df646fca6419a1fc4ef61b89d9 8086a37957a2ad182ae9562e10e24687edc2947e

I also checked the full diff, mount failure paths, and test registration. I downloaded the author-supplied test archive and verified its SHA-256 against the PR description; the runs below are still author-provided, not runs I performed.

Results

Check	Result
Diff whitespace and conventions/template	Passed.
ABI audit	passed_with_warnings; lexical warnings concern exports outside this PR's changed paths.
Test-quality heuristic	One possible-unregistered warning; manual inspection confirmed registration in libSceSaveData.native/CMakeLists.txt.
Merge preview	Clean against the PR base.
Author-provided Linux qualification	Archive SHA-256 matched eb8db3423918202bceaa39eb8ba550d1bf469d715d9f981c6ddeedfc7dd12621. Its logs report 18/18 baseline and 18/18 candidate passes across six suites, a negative-control failure using the baseline module with the new test, and a permission probe that aborted 3/3 on baseline and passed 3/3 on candidate. This was a module/suite qualification, not a clean whole-repository build.
Hosted CI	All listed jobs were queued at final refresh.

Unedited local check output

The no-build precheck reported a tooling-related failure because its isolated worktree lacks the workspace-local ABI and test-quality scripts; both checks were run manually. The ABI audit produced only lexical warnings for exports outside the changed paths.

Test-quality heuristic:

{"findings": [{"path": "core/libs/tests/GuestSaveDataMountStatus.cpp", "line": 1, "kind": "possible-unregistered-cpp-test", "message": "Changed C++ test source path was not found literally in CMake files; inspect registration, generated source lists, and workflow coverage."}], "negative_control_candidates": [], "limits": ["Heuristics can miss weak assertions and can flag valid tests; inspect every finding.", "A test that passes against base is not by itself proof that the PR test is inadequate.", "Negative controls are only candidates until the test actually runs against base and fails by assertion rather than import/setup error."]}

Conventions and merge preview:

All conventions checks passed.
e0a3bd33285606d39994c8a5f3b154df986e87dd

Negative control

Assertion	Change to code under test	Result
Failed mount returns an error without publishing state	Author's archive reports running the new test with the baseline module	The new assertions failed on baseline; candidate tests passed.

Findings

Blocking

None found in the reviewed source and supplied Linux qualification.

Non-blocking

The archive documents Linux behavior only; firmware-specific error mapping remains explicitly unverified in the PR's debt entry.

Sources

• Mount3 error handling and delayed publication at the reviewed commit.
• Failure-state regression cases at the reviewed commit.
• Recorded firmware error-mapping limitation.

Not verified

I did not independently run or build the C++ tests, run Windows/macOS, test a game, or verify the corresponding PS5 firmware error mapping. Hosted CI remained queued.

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
PR 12853	core/libs/prx/libSceSaveData.native/Export.cpp L334-335
PR 13622	core/libs/prx/libSceSaveData.native/Export.cpp L665-667

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 10674	core/libs/prx/libSceSaveData.native/Export.cpp docs/dev/debt/libSceSaveData.native.md

[adriadam10]

The Windows Build fails on this head: guest_savedata_mount_status aborts with FATAL: C++ exception filesystem_error raised ... libSceSaveData.native.prx+0x1459a has no unwind data (run 41095969559).

• core/libs/prx/libSceSaveData.native/Export.cpp:689-693: the try/catch in sceSaveDataMount3 cannot work on Windows, because the PRX has no unwind data and the exception cannot be delivered to the handler. The new CheckHostMountFailures case (a file blocking _sd) hits it.
• Fix: do not throw inside mount3. Use the std::error_code overloads for is_directory, exists, absolute and create_directories (Export.cpp:651, 663-665) and return SAVE_DATA_ERROR_INTERNAL when ec is set, then drop the try/catch. The same rule applies to blocks_path/replace_file if either can throw.

The rest (sidecar written before publishing alias/slot, cleanup of the new empty directory, trace reading output only on success) looks right; Linux is not the problem here.

[JJenkx]

Addressed in 2606c33532a52416ed993f56814441e7d80f1a96: removed the catch and switched Mount3's filesystem probes, directory creation and mounted-directory identity checks to error-code handling. A missing save still follows normal CREATE/NOT_FOUND behavior; other host errors return INTERNAL before publishing mount state. The sidecar replacement helper already uses error-code directory creation and native/nonthrowing rename.

Rebuilt the full module and six SaveData suites on Linux: 18/18 baseline and 18/18 change passes; the unprivileged permission regression passes 3/3 on the change. Updated the description with exact-head logs and a clean merge check against main 4bc9e779aa7a. Windows validation is requested and pending; I am not claiming a Windows pass yet.

[JJenkx]

Windows rerun completed on exact head 2606c33532a52416ed993f56814441e7d80f1a96: all six guest SaveData suites pass with zero skips. In the same isolated build, prior head 8086a37957a2ad182ae9562e10e24687edc2947e reproduces the reported filesystem_error / PRX “has no unwind data” abort in mount status; the other five suites pass.

Updated the description with full Windows build/test logs and JUnit results, SHA-256 75e001074bd44a5c82564673b1fba0b791532074a6b32ff5beda9149e06c297e (download verified). Focused module/suite qualification only; no whole-project or game/GPU claim.

[adriadam10]

Checked that the try/catch is gone and mount3 no longer calls any throwing filesystem overload: probes, absolute, create_directories and the mounted-directory identity check take std::error_code, and replace_file is non-throwing. The test pins result, alias and slot preservation for both blocking cases plus recovery, and fails on the old code; merged.